Conventional automatic speech recognition (ASR) systems typically perform well in converting speech to text. However, in some cases, the same speech sounds map to different words with different meanings. These types of words are commonly known as heterographs. In these situations, conventional ASR systems will either choose one of the words matching the received speech at random or request input from the user to clarify which word to use. This results in either the wrong output or frustration for the user.